GCP Data Practitioner key terms

A BigQuery table whose data stays in Cloud Storage, BigLake, or another source rather than being copied into BigQuery storage, so queries read the source files each time they run.

A BigQuery-managed table over open-format files in Cloud Storage that adds fine-grained access control, metadata caching, and performance features beyond a plain external table.

A Google Cloud service for discovering, cataloging, governing, and monitoring data across lakes, zones, and assets, including BigQuery datasets and Cloud Storage buckets.

A record of where data came from, how it moved, and which transformations produced it, used for debugging, impact analysis, and audit.

A searchable inventory of datasets, tables, and their metadata and business descriptions, so analysts can find trustworthy data without knowing its physical location.

A statistical summary of a dataset such as null counts, distinct values, min and max, or value distribution, used to detect quality problems before analysis.

A declared expectation about data, such as uniqueness of a key or a non-null constraint, that a pipeline or profiling job checks and reports on.

A change in the shape of incoming data, such as added, renamed, or retyped fields, that can break downstream loads or queries if it is not handled.

A load or streaming option where the service infers column names and types from the incoming records instead of requiring a predefined schema.

A query optimization where the engine skips partitions that cannot match the filter, reducing bytes scanned and cost for time- or value-partitioned tables.

A table physically split into segments by a column such as a date or ingestion time, so queries and retention operations can target individual segments.

Sorting and co-locating table rows by one or more columns so that queries filtering on those columns read fewer blocks of data.

The amount charged for a BigQuery query, typically based on bytes scanned in on-demand pricing or on slots consumed under capacity pricing.

A validation run that checks a query or job for syntax and permissions and reports estimated bytes processed without actually executing it or billing for data.

A precomputed, automatically refreshed result set stored by the warehouse that the engine can rewrite queries to use, avoiding repeated expensive aggregation.

A directed acyclic graph: a set of tasks connected by dependencies with no cycles, used to express the order in which pipeline steps must run.

A managed Apache Airflow service for authoring and scheduling workflow DAGs, including operators that interact with Google Cloud services.

A service for building SQL-based data transformation workflows in BigQuery, with version control, dependency graphs, assertions, and scheduled runs.

A pipeline designed so that re-running a task or replaying a batch produces the same final state, which makes retries and backfills safe.

Reprocessing historical time ranges of data, often after a logic fix or a new pipeline is deployed, to bring older partitions up to date.

A separate destination where messages that repeatedly fail processing are stored, so they can be inspected and retried without blocking the main stream.

A named resource attached to a topic that delivers published messages to subscribers, with options for push or pull delivery and retention.

A named channel to which publishers send messages; each topic can have multiple subscriptions that each receive their own copy of the stream.

Grouping unbounded streaming data into finite time or session windows so that aggregations can be computed over a bounded set of events.

A progress signal in stream processing indicating that events earlier than a given point are believed to have arrived, which triggers window results.

Events that reach the processing system after their window has already been emitted, requiring handling such as allowed lateness or reprocessing.

A delivery guarantee where each input record affects the output exactly one time, even across retries or worker failures, avoiding duplicates and losses.

A managed service for running Apache Spark and Hadoop clusters on Google Cloud, often used for batch processing and migration of existing big-data jobs.

A managed, code-free data integration service for building pipelines with a visual interface, including connectors and transformation operators.

Access control that restricts which users or groups may read specific columns of a table, often implemented with policy tags in a data catalog.

A label attached to a column in a data catalog that maps to an IAM-controlled taxonomy, enforcing who can read that column's values.

A filter applied by the warehouse that returns only rows a given user is permitted to see, enforced inside the query engine rather than in the client.

A view granted access to a source dataset so that users can query the view's results without being granted direct access to the underlying tables.

A unit of computational capacity used to run BigQuery queries; capacity pricing reserves slots, while on-demand pricing draws from a shared pool.

A committed allocation of BigQuery slots to a project, folder, or organization, with optional autoscaling and idle-slot sharing.

A technique that captures inserts, updates, and deletes from a source system so they can be replicated to a warehouse with low latency.

A serverless change data capture and replication service that streams changes from databases such as MySQL and PostgreSQL into BigQuery or Cloud Storage.

A managed service that schedules recurring copies of data into BigQuery or Cloud Storage from sources including SaaS apps, databases, and other clouds.

A managed service for moving or syncing large object sets between Cloud Storage buckets, other clouds, or on-premises sources on a schedule.